ocminer/keryx-miner-supr v0.10.5

on GitHub

Diagnostic release. No consensus or protocol changes — a drop-in upgrade for v0.10.3/0.10.4 miners. The in-process llama engine is unchanged from v0.10.4.

What this fixes

Previously, a manually-placed model.gguf that the miner could not adopt (wrong path/filename, or a truncated/incomplete download) produced only a generic, repeating "preparing models (staging/verifying files)" line with no reason. The model never earned its .ok sentinel, so it never counted as ready and the OPoI gate suspended mining indefinitely — with nothing in the log explaining why.

Changes

• Names the exact reason a model was rejected — absent, not a parseable GGUF, or incomplete/truncated — on-disk X < required Y for N tensors.
• Prints the exact expected path the miner is looking for, and the required <model-dir>/<dir_name>/model.gguf layout.
• Periodic per-model staging status (~once/min) while stuck in "preparing", listing every lineup model as READY / complete but not yet adopted / NOT ready — <reason>.

If a hand-copied model won't go ready, the log now tells you which model it wants and why the file on disk was refused (most commonly: a truncated download, or the wrong folder/filename).

Builds

Each line ships HiveOS (*.tar.gz) and Linux (*-linux-x86_64.tar.gz) packages with both AVX and noavx llama CPU backends. AMD and Windows builds are the *-amd-* / *-windows-* assets. Checksums: SHA256SUMS-0.10.5.txt (NVIDIA Linux/HiveOS).